Re: equilibration of solvent box and pressure

From: Peter Freddolino (
Date: Mon Oct 06 2008 - 15:24:08 CDT

Hi Hiro,

Hiroaki Maekawa wrote:
> 1) Do I need to plot PRESSURE or PRESSAVG?
PRESSAVG is generally going to be more useful. The number of outputs is
identical because pressavg is a local average; if you output energies at
step 100, 200, etc., PRESSURE contains the values at steps 100, 200,
etc., whereas PRESSAVG at step 100 has the average pressure over steps
1-100, at 200 has the average over steps 101-200, etc.
> I just wonder the latter is the average between energy outputs
> according to the manual,
> but the number of printed data is same for PRESSURE and PRESSAVG.
> Why is it?
> Is it true that the PRESSURE value doesn't make any sense if multi
> time step is
> used to calculate different forces?
PRESSURE values don't make any sense if multiple timestepping is used
*unless* the frequency of energy output is an integer multiple of all of
the step frequencies used for different forces. If the energy output
frequency is appropriate, you'll get proper values reflecting the forces
being applied.
PRESSAVG also properly takes this into account.
> In addition, is it better to take average of all of 2 ns data or just
> final 500 ps is enough
> to check whether the average pressure is close to the target value?
Usually you'll want to take an average over some period of time *after*
the initial equilibration period, and verify that your conserved
quantities are stable and have an average within the statistical error
of the values they should take. You'll also probably note in these
calculations that pressure tends to have quite large fluctuations.
> 2) Does just a much longer simulation lead to the average value or
> is there any good technique to shorten the equilibration time?
> I just would like to know how long the equilibration needs to be
> continued usually.
The best way to have a short equilibration time is to start with a
system that is near equilibrium; for example, if you're having pressure
issues, starting from an appropriately sized unit cell will help. In the
case of a pure, homogeneous liquid simulation it would be very unusual
to need an equilibration longer than what you've already done. A longer
simulation should eventually give you an average of exactly the pressure
your barostat is set to, but this can take quite a long time due to the
magnitude of the deviations observed in instantaneous pressure.

> Thank you for your answers.
> best
> Hiro

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:21:22 CST